Ziwei Brief

Scoring · ruleset palace-score-v1

How a palace score is computed

A palace score is a sum, not a verdict. Each term is printed next to the number, so you can add them up yourself and land on the same result the engine did. This page is the rule, the arithmetic on one real chart, and the audit code that lets you check the arithmetic was not quietly changed between the sample and your copy.

For entertainment and self-reflection. Not medical, legal, financial, or psychological advice. We do not predict the future; you decide.

The five steps

  1. Every palace starts at the same base: 50. A score is not a comparison between palaces on your chart, so there is no normalising step and no percentage.
  2. Each star in the palace contributes a fixed amount, decided by its brightness — Bright +12, Neutral +4, Dim -8, and unmarked +2 when the source gives no brightness at all.
  3. A transformation on that star adds or subtracts again, on top of the brightness term. This is the largest single term in the whole rule: Hua Ji -14.
  4. Minor stars, malefics and support stars add their own stated amounts, and an empty palace pays a fixed penalty for borrowing the opposite palace's main stars (-5).
  5. The terms are added, and the total is clamped to 0–100. A sum of 112 is reported as 100; a sum of -6 is reported as 0. The clamp is visible on the page where it happens — this chart has none.

One palace, term by term

The chart below is the demonstration chart the free sample brief is rendered for — 12 May 1990, 08:30 · Shanghai. It is synthetic and belongs to nobody. Terms are listed in the order the brief prints them: heaviest first, so the two or three things that actually moved the number are at the top. The running total is what you would write down if you did the sum by hand.

The palace with the most terms on this chart is the Travel Palace — 14 terms, distance field, empty, so it borrows the opposite palace's main stars.

Running total for the Travel Palace, from the base score to the final clamped score.
TermChangeRunning total
Base—50
Tai Yin · bright · borrowed+1262
Tai Yang · Hua Lu · borrowed+1072
Di Kong-963
Tai Yang · dim · borrowed-855
Ling Xing-748
Tuo Luo-741
Empty palace, borrows the opposite-536
Tai Yin · Hua Ke · borrowed+541
Tian Yue+344
You Bi+347
Zuo Fu+350
Kong Wang050
Tian Cai050
Tian Kong050
Sum before clamp—50
Scoreno clamp needed50 · Tight

Two things in that table look like mistakes and are not. A star carrying a transformation appears twice — once for its brightness, once for the transformation — because they are two separate terms with two separate weights. And several stars contribute nothing at all: they are still printed, at zero, because a term that is silently dropped is a term the reader cannot audit.

Nothing about this palace is hidden in a model. If you disagree with the result, you can see exactly which term you disagree with — usually the size of one delta, not the sum. The same applies to the twelve scores in the free sample brief, which prints the same table for sixteen pages.

The terms, in full

This is the whole rule for ruleset palace-score-v1. Every palace on every chart is scored from these numbers and nothing else — no randomisation, no time of day, no model call.

Main stars, by brightness

  • Bright+12
  • Neutral (the common case)+4
  • Dim-8
  • No brightness in the source data+2

The four transformations

  • Hua Lu+10
  • Hua Quan+6
  • Hua Ke+5
  • Hua Ji-14
  • Hua Ji landing on the Life, Wealth, Career or Spouse palace — an extra term-4

Malefics

  • Qing Yang · Blade-7
  • Tuo Luo · Spiral-7
  • Huo Xing · Fire Star-7
  • Ling Xing · Bell Star-7
  • Di Kong · Earthly Void-9
  • Di Jie · Earthly Robbery-9

Support stars

  • Zuo Fu · Left Support+3
  • You Bi · Right Support+3
  • Tian Kui · Heavenly Noble+3
  • Tian Yue · Heavenly Halberd+3
  • Wen Chang · Literature+3
  • Wen Qu · Eloquence+3
  • Lu Cun · Wealth Keeper+3
  • Tian Ma · Heavenly Horse+3
  • Any other minor star not named above+1
  • Any other support star not named above — scored at zero, still printed0

Structure

  • Empty palace borrowing the opposite palace's main stars-5
  • Clamp applied to the final sum0–100

The four bands

Bands are thresholds on the absolute score, not a rank against other charts. They exist so a number can be read without arguing about whether 64 is good.

Band thresholds and what each band means.
BandScoreReading
Clear70–100The area runs on its own terms. Contact with it tends to return more than it costs.
Workable55–69Usable, with a cost attached. Nothing is blocked; something is being paid for.
Tight42–54Movement here takes more out of you than it gives back. It is not a prohibition.
Blocked0–41The heaviest band. Contact is expensive; that is a statement about cost, not outcome.

The audit code

Everything above is a claim that the numbers come from the birth data and the ruleset, and from nothing else. A claim like that is worth exactly what it costs to check, so the brief prints a short code that binds the two together.

The code is the first eight hex characters of a SHA-256 hash over a canonical text form: one line naming the ruleset, then one line per palace — id, score, band — sorted by id, so the order the palaces happen to be listed in cannot change the result. Birth data, order reference, render time and prose are deliberately excluded: anything that varies between two renders of the same reading would make the code useless for comparison.

Hashed text for the chart above

ruleset=palace-score-v1
career=35=Blocked
children=62=Workable
inner=65=Workable
life=72=Clear
parents=54=Tight
partnership=44=Tight
peers=65=Workable
property=74=Clear
siblings=80=Clear
travel=50=Tight
vitality=71=Clear
wealth=62=Workable

SHA-256 of that text, first eight characters: Audit FFCE-6952

Enter the same birth data at the verification page and you will get the same code back — that is the whole check. If it differs, either the input differed or the rules changed, and the ruleset id on both documents tells you which. Eight characters is a fingerprint, not a signature: it catches a different ruleset or a different score, and it does not prove who produced the file.

All twelve, on this chart

The full board, so the worked example above is not a cherry-picked one. Term counts are shown because a score built from two terms and a score built from nine are not equally informative, even when the numbers are close.

All twelve palaces for the demonstration chart, with score, band and number of terms.
PalaceScoreBandTerms
Life Palace72Clear5
Siblings Palace80Clear8
Spouse Palace44Tight6
Children Palace62Workable7
Wealth Palace62Workable5
Vitality Palace71Clear7
Travel Palace50Tight14
Friends Palace65Workable6
Career Palace35Blocked3
Property Palace74Clear6
Inner Life Palace65Workable7
Parents Palace54Tight5

What the score is not

  • Not a measurement of a person. It is a count of the terms a model finds in a chart. The model is public, which makes the count checkable; it does not make the count true.
  • Not stable across rulesets. A different set of deltas scores the same birth data differently. That is not a bug to be papered over — it is why the ruleset id is printed on every document and hashed into every code.
  • Not a prediction. A band says what contact with an area tends to cost, given the terms in it. It does not say what will happen, and it cannot tell you whether to take a job.
  • Not comparable between people as a ranking. Two charts can both show 61 in the Career palace without the two careers being equivalent; the absolute scale is there to be read on its own, not as a league table.

Common questions

Is the score arbitrary?

No — it is fixed arithmetic. Each star and transformation carries a stated delta, the deltas are added to a base of 50, and the total is clamped to 0–100. You can reproduce any palace's score by hand from the terms printed beside it. What is a judgement call is the size of the deltas, not the sum: those weights are our choice, they are published here, and a different ruleset would move the numbers.

Why 0–100 and not a percentage?

Because it is an absolute scale, not a ranking against other people. 70 means the same thing on every chart. A percentage would invite the reading "better than 70% of charts", which the engine never computes and could not support.

What stops the scores from changing after I buy?

The ruleset id and the twelve score/band pairs are hashed into an audit code printed on the brief. Recompute the same birth data at /verify and you get the same code back. A rules change means a new ruleset id, which means a different code — visibly, not silently.

Start here

Read next